Bug 179769

Created attachment 327064 [details] Patch

Comment on attachment 327064 [details] Patch Nice!

Comment on attachment 327064 [details] Patch Clearing flags on attachment: 327064 Committed r224921: <https://trac.webkit.org/changeset/224921>

All reviewed patches have been landed. Closing bug.

<rdar://problem/35590236>

We shouldn't be doing this. These tests are primarily useful when shared with other teams, so requiring a complicated harness to run them is unacceptable.

Re-opened since this is blocked by bug 179813

Reopening for further discussion.

I'm afraid I don't really understand your objection; could you elaborate?

Given the discussion in webkit-dev, I think we can proceed with this patch: - removed tests are clearly conformance tests and not regression tests - added tests are slightly better than removed tests and will be kept up to date

> I'm afraid I don't really understand your objection; could you elaborate? When there is a problem in graphics frameworks below WebKit, I can easily put the test in a zip archive, and ask the team working on the framework to debug what's going on. They can edit the test locally as they see fit to simplify it as part of their investigation. Requiring a full WebKit checkout and complicated scripts that may or may not even work in their OS build will make getting fixes for such problems substantially trickier.

Creating a zip archive is not straightforward either. It still requires to get all the necessary resources available which might be tricky. WPT tests are available at https://w3c-test.org.

One cannot edit the w3c-test.org copy, which is important for debugging.

(In reply to Alexey Proskuryakov from comment #13) > One cannot edit the w3c-test.org copy, which is important for debugging. Is a webarchive somehow hackable? If so, it should be fairly easy and less error-prone than a zip archive to create a webarchive from an existing webpage, send that webarchive and edit it.

Given the discussion in webkit-dev, we should probably reland this patch. Please let me know if there is any objection.

(In reply to youenn fablet from comment #15) > Given the discussion in webkit-dev, we should probably reland this patch. > Please let me know if there is any objection. @Youenn: I'm not sure which part of the conversation makes you think we should, but as I understand the situation is stalled until we improve runtime performance of WPT tests on WebKit. From my point of view, sharing the tests with the rest of the web platform community (to keep it up-to-date, being able to compare results, etc) is more important than making it convenient for Apple to work internally on proprietary code outside WebKit. I wonder, if maybe we should still import the new tests and let each port maintainers decide which version of the tests they want to skip in TestExpectations?

> @Youenn: > > I'm not sure which part of the conversation makes you think we should, but > as I understand the situation is stalled until we improve runtime > performance of WPT tests on WebKit. From webkit-dev, it seems fine to continue importing WPT tests, including these 2d tests. The main discussion point is whether we should remove the corresponding WebKit tests. Since the existing WebKit 2dcontext tests are not regression tests but conformance tests, I think it is fine to remove them as well.

> Is a webarchive somehow hackable? It's not. Definitely not within easy reach of someone who works on lower level frameworks. > From webkit-dev, it seems fine to continue importing WPT tests, including these 2d tests. Let's give everyone who just got back from the Thanksgiving holiday (including myself) catch up on the discussion. My feeling is that test stability and performance is more important for WebKit right now than adding more tests. This is obviously a very general statement, we can have lots of special cases.

@David @Simon: Instead of having 3 separate bugs for this issue, we should probably merge all of them into one. Can we mark bug 193579 and bug 193042 as duplicate of this one?